chore(ci): reduce SDK duplication and extend image checks #792

Follow-up to #787. Keep that PR focused on the Server 1.7 release fixture, necessary initialization and small cache corrections. The following changes should be implemented separately.

Reuse candidate SDK preparation within one workflow run

Consumer jobs currently repeat preparation of the same locked Server SDK, and the Hubble fixture also builds a distribution. Evaluate producing the SDK artifacts, Server archive and provenance manifest once, then restoring them into isolated consumer directories.

• Preserve source identity, artifact hashes and locally built SDK provenance checks.
• Keep consumer behavior tests and independent required/advisory cancellation intact.
• Cache third-party dependencies separately; do not substitute stale HugeGraph SDK artifacts.
• Compare total runner time and the critical path before adopting the change. Do not introduce cross-run success reuse or a new scheduling framework.

Exercise the final Loader image and example Compose

The image checks cover the Loader executable entry point and Hubble readiness, page and static resources. They do not yet demonstrate a real import through the final Loader image or execute the Loader example Compose.

• Use the image built from the current checkout, including when running the actual example Compose with an explicit image override.
• Import a tiny dataset through the published Loader entry point, then query Server to verify the result.
• Keep startup bounded and clean up containers on success and failure.
• Retain existing image identity checks; avoid duplicating the full consumer or Hubble browser suites.

These are advisory CI follow-ups, not additional merge requirements. They may be delivered in separate small PRs.